Safety is critical to broadening the real-world use of reinforcement learning. Modeling the safety aspects using a safety-cost signal separate from the reward and bounding the expected safety-cost is becoming standard practice, since it avoids the problem of finding a good balance between safety and performance. However, it can be risky to set constraints only on the expectation neglecting the tail of the distribution, which might have prohibitively large values. In this paper, we propose a method called Worst-Case Soft Actor Critic for safe RL that approximates the distribution of accumulated safety-costs to achieve risk control. More specifically, a certain level of conditional Value-at-Risk from the distribution is regarded as a safety constraint, which guides the change of adaptive safety weights to achieve a trade-off between reward and safety. As a result, we can compute policies whose worst-case performance satisfies the constraints. We investigate two ways to estimate the safety-cost distribution, namely a Gaussian approximation and a quantile regression algorithm. On the one hand, the Gaussian approximation is simple and easy to implement, but may underestimate the safety cost, on the other hand, the quantile regression leads to a more conservative behavior. The empirical analysis shows that the quantile regression method achieves excellent results in complex safety-constrained environments, showing good risk control.

Invalid action masking is a practical technique in deep reinforcement learning to prevent agents from taking invalid actions. Existing approaches rely on action masking during policy training and utilization. This study focuses on developing reinforcement learning algorithms that incorporate action masking during training but can be used without action masking during policy execution. The study begins by conducting a theoretical analysis to elucidate the distinction between naive policy gradient and invalid action policy gradient. Based on this analysis, we demonstrate that the naive policy gradient is a valid gradient and is equivalent to the proposed composite objective algorithm, which optimizes both the masked policy and the original policy in parallel. Moreover, we propose an off-policy algorithm for invalid action masking that employs the masked policy for sampling while optimizing the original policy. To compare the effectiveness of these algorithms, experiments are conducted using a simplified real-time strategy (RTS) game simulator called Gym-$\mu$RTS. Based on empirical findings, we recommend utilizing the off-policy algorithm for addressing most tasks while employing the composite objective algorithm for handling more complex tasks.

Multiagent reinforcement learning (RL) training is usually difficult and time-consuming due to mutual interference among agents. Safety concerns make an already difficult training process even harder. This study proposes a safe adaptive policy transfer RL approach for multiagent cooperative control. Specifically, a pioneer and follower off-policy policy transfer learning (PFOPT) method is presented to help follower agents acquire knowledge and experience from a single well-trained pioneer agent. Notably, the designed approach can transfer both the policy representation and sample experience provided by the pioneer policy in the off-policy learning. More importantly, the proposed method can adaptively adjust the learning weight of prior experience and exploration according to the Wasserstein distance between the policy probability distributions of the pioneer and the follower. Case studies show that the distributed agents trained by the proposed method can complete a collaborative task and acquire the maximum rewards while minimizing the violation of constraints. Moreover, the proposed method can also achieve satisfactory performance in terms of learning speed and success rate.

Energy optimization is crucial for unmanned aerial vehicle (UAV) path planning, particularly in complex wind-field environments. Most existing path-planning algorithms rely on simplified energy consumption models, which often fail to adequately capture the effects of wind fields. To address this limitation, a wind-adaptive lifelong planning A* algorithm (WA-LPA*) is proposed for energy-aware path planning in dynamic wind environments. WA-LPA* constructs a composite heuristic function incorporating wind-field alignment factors and integrates a hierarchical height-aware optimization strategy. Meanwhile, an adaptive replanning mechanism is designed based on the change characteristics of the wind field. Simulation experiments conducted across representative scenarios demonstrate that, compared to conventional algorithms that neglect wind-field effects, WA-LPA* achieves energy efficiency improvements of 5.9–29.4%.